How to Write Google Ads: A Message-Matched RSA Workflow

To write Google Ads for a video sales letter, start with one closely related query cluster and define the strongest promise that both the ad and destination can support. Then assign each responsive search ad (RSA) asset a clear role and check that plausible combinations lead naturally into the video sales letter (VSL).

Treat the RSA as a message system, not a bag of interchangeable slogans. Google’s diagnostics can identify limited or repetitive assets, but they cannot prove that an ad will convert.

Group queries by the searcher’s immediate objective, not merely by shared words. Informational, comparison, and purchase-oriented queries may use similar language while requiring different messages.

Record these inputs before writing:

• The searcher’s likely objective
• The offer being presented
• The qualified audience
• The destination’s opening promise
• Verified facts that support the message
• Claims the ad must not make
• The primary business metric you intend to evaluate

The message boundary is the strongest accurate promise that both the ad and destination can support. It prevents individual assets from escalating the offer when Google assembles them in different combinations.

Hypothetical query-to-promise map

Suppose a company sells bookkeeping software for independent consultants.

“Never make another bookkeeping mistake” exceeds the boundary unless the company can substantiate that absolute outcome. “Organize consulting income and expenses in one workflow” may fit if the product supports those tasks.

Do not assume that every query containing “bookkeeping” belongs in the same group. A person looking for a bookkeeping course has a different objective from someone seeking software.

Google recommends creating varied RSA assets, but describes Ad Strength as guidance rather than proof that a particular message will produce business results. Asset diversity is useful only when every option remains inside a coherent message boundary (Google Ads guidance).

Useful headline roles include:

• Query recognition: Reflect the problem or category.
• Audience qualification: Identify whom the offer serves.
• Promise: State a bounded benefit.
• Mechanism: Explain how the offer approaches the problem.
• Proof: Present a specific, verified credibility fact.
• Next step: Tell the searcher what to inspect or do.

Here, a mechanism is simply the process, feature, or approach through which the offer addresses the problem. Descriptions can provide fuller explanations, qualifications, conditions, and transitions to the destination.

Selected transcripts in the supplied proprietary sample followed a sequence of recognition, tension, promise, and explanation. Other sampled transcripts opened with a counterintuitive reframe that changed how viewers might categorize the problem. These are structural observations from a limited sample, not evidence that either sequence improves performance.

A simple role map looks like this:

Query recognition ─┐
Qualification ──────┤
Promise ────────────┤
Mechanism ──────────┼─> Verified message boundary ─> VSL opening
Proof ──────────────┤
Next step ──────────┘

Each asset should contribute something distinct. If an asset requires a different audience, offer, mechanism, or level of certainty, move it to another message group or test.

Start with literal headlines that make the query-to-offer relationship clear. Add reframes, proof, and mechanism language only after establishing that basic relevance.

Hypothetical headline set

The proof example illustrates a common defect: a plausible statement is not automatically an authorized statement. Concise credibility markers appeared near the first meaningful promise in several sampled transcripts, but adapting that placement still requires verifiable brand facts.

A question or reframe can make the situation recognizable without inflating the stakes. “Spending Client Time on Books?” describes an observable tension. “Is Bookkeeping Destroying Your Business?” introduces a much stronger claim that the offer may not be able to support.

Check every headline independently:

1. Can the offer support its literal meaning?
2. Is it accurate without the other headlines?
3. Could another asset turn it into an absolute claim?
4. Does the VSL address the same idea near its opening?
5. Is every number, comparison, credential, or outcome backed by a current record?

Do not rely on a description to rescue an unsupported headline. That description may not appear beside it.

A description should extend the message rather than repeat a headline. Give each description a primary role:

• Explanation: Clarify what the product or process does.
• Qualification: Identify the relevant audience, condition, or limitation.
• Credibility: Add an accurately bounded and verified fact.
• Next step: Preview what the visitor will find after clicking.

Hypothetical descriptions

This description explains the mechanism. “Supported accounts” also avoids suggesting universal compatibility.

This version qualifies the audience and sets a concrete destination expectation.

Some narratives in the sampled corpus placed their explanations inside a discovery-oriented journey. You can adapt the progression from problem to practical question to explanation. Do not invent a founder experience, customer result, expert endorsement, or breakthrough to make it more dramatic.

Read each description beside each promise headline. Flag any pairing that changes a conditional benefit into a guarantee or makes a feature sound universally available.

Message match does not require repeating the same words. It requires continuity across four elements:

1. Problem: The VSL addresses the situation implied by the query.
2. Promise: The VSL retains the ad’s level of certainty.
3. Mechanism: The approach introduced in the ad appears early enough to orient the viewer.
4. Qualification: Important audience and offer conditions remain visible.

Hypothetical continuity audit

When the audit reveals a break, revise the unsupported side. If “error-free” lacks substantiation, remove it from the VSL rather than raising the ad claim to match.

Apply the same discipline to convenience claims. One sampled corpus pattern connected format to routine and usability rather than treating convenience as proof of superior performance. For software, the equivalent is describing the verified workflow without claiming that convenience guarantees better records or business results.

Inspect combinations that could become misleading even when each asset appears acceptable alone. Promise-plus-proof and promise-plus-mechanism pairings deserve particular attention.

For each plausible combination, ask:

• Does it answer the query?
• Do all assets describe the same offer?
• Is the audience qualification preserved?
• Does the mechanism exist on the destination?
• Do two modest claims combine into an absolute claim?
• Can the VSL continue the message without changing its meaning?

Google describes the RSA combinations report as diagnostic. Served combinations depend on real-time signals, and recreating a reported combination does not guarantee identical performance (Google Ads combinations guidance). Use the report to identify semantic gaps, not to collect supposedly repeatable templates.

Pinning an asset may preserve an essential qualification or required statement, but it also limits which combinations Google can assemble. Treat pinning as a message-control decision whose effect must be evaluated in the campaign’s actual context.

A coherent RSA is ready to test; it is not proven effective. Define one principal message variable and one primary business metric before interpreting results.

Google advises advertisers to use controlled experiments and warns that simultaneous changes make attribution ambiguous (Google Ads Experiments guidance). That guidance does not provide a universal sample-size rule for every campaign. An early numerical difference is therefore not enough to establish a reliable result.

Hypothetical single-variable test brief

• Question: Does consultant-specific framing lead to more qualified demo requests than general small-business framing?
• Principal variable: Audience-qualified language versus general category language
• Primary metric: Qualified demo requests
• Held constant: Offer, destination, query cluster, geography, bidding approach, and conversion definition
• Secondary diagnostic: Click-through rate
• Interpretation limit: A click-through-rate difference alone does not establish better lead quality or downstream revenue

Ad position, competition, targeting, bidding, and user context can all influence response. Do not interpret click-through rate as a copy-only verdict, and do not assume that message continuity improves VSL completion or revenue until those outcomes are measured directly.

Complete this worksheet before testing the ad:

QUERY CLUSTER
Primary intent:
Representative queries:
Excluded or materially different intents:

MESSAGE BOUNDARY
Offer:
Qualified audience:
Strongest verified promise:
Verified mechanism:
Available proof:
Claims not supported:

HEADLINE ROLES
Query recognition:
Audience qualification:
Promise:
Mechanism:
Proof:
Next step:

DESCRIPTION ROLES
Explanation:
Qualification:
Credibility:
Next step:

VSL CONTINUITY
Opening problem matches:
Promise keeps the same certainty:
Mechanism appears early:
Qualification is preserved:
Unsupported escalation found:

COMBINATION REVIEW
Potentially ambiguous pairing:
Potential absolute claim:
Asset needing revision:
Pinning required and why:

TEST BRIEF
One principal variable:
One primary business metric:
Elements held constant:
Interpretation limits:

The worksheet is complete when every asset has a defined job, plausible combinations remain inside the verified boundary, and the VSL can continue the message without escalation.

A clear next step can close the path without invented urgency. Sampled closing structures directed viewers toward concrete purchase information, but conditions such as price, guarantees, discounts, inventory, and shipping still require current verification. Tell visitors what they can inspect without implying terms the destination cannot confirm.

Daily Intel is designed to research ads, VSLs, transcripts, offer context, and funnel paths in one workflow. Use that connected view to compare how messages move from query to ad to destination, while keeping sampled structural observations separate from evidence of performance.